Despite the adoption of "sickest first" liver transplantation, pretransplant death remains common, and many early deaths occur despite initially low Model for End-stage Liver Disease (MELD) scores. From 1997-2003, we studied 507 cirrhotic United States veterans referred for consideration of liver transplantation to identify additional predictors of early mortality. Most of the patients were male (98%) with cirrhosis caused by hepatitis C and/or alcohol (88%). Data for 296 patients referred prior to February 27, 2002 (training group), were analyzed; findings were validated in 211 patients referred subsequently (validation group). In the training group, 61 patients (21%) died within 180 days without transplantation; their median initial MELD score was 21. MELD score, persistent ascites, and low serum sodium (<135 meq/L) were independent predictors of early mortality. In patients with a MELD score of less than 21, only low serum sodium and persistent ascites were independent predictors of mortality; for MELD scores above 21, only MELD was independently predictive. Prognostic significance of persistent ascites and low serum sodium for low MELD score patients was confirmed in the validation group. Risk varied continuously with worsening hyponatremia. Modifying MELD, by including points for persistent ascites and low serum sodium, improved prediction of early pretransplant mortality in low MELD score patients. In conclusion, persistent ascites and low serum sodium identify patients with cirrhosis with high mortality risk despite low MELD scores. Ascites, hyponatremia, and other findings indicative of hemodynamic decompensation merit further prospective study as prognostic indicators in patients awaiting liver transplantation, and should be considered in setting minimal listing criteria.